Questions from a new folder: How long does a large SMP work unit usually take?

[andper10]

So now that I am not paying for my electricity and it is getting colder outside (and therefore colder inside too), I figured I might as well set up my PC to fold when I'm not using it.
I installed the latest Beta SMP Client to run on my Q9550. After following a ridiculous amount of tedious command prompt configuration directions from the F@H website, it seems to be installed per their instructions. However, I am wondering how long it normally takes to process a 500000 step WU on a Quad Core CPU. Right now it has taken about 3 hours to get 3% (15000 steps) done. Does this seem about right for a stock Q9550 CPU? (I plan to reinstate the overclock in my sig in the next few days now that I have a reason to and now that it is cooler inside.) So it looks like I will have to run the computer for about 14.5 hours per day to meet the 7 day deadline for this WU. (I will try to replace the led fans so I can leave it running 24/7 without it being bothersome.)

P.S. I also installed the GPU client to run on my 4870 and it seems to be working well. The first WU was 10000 steps and it took around 4 hours to complete. I am thinking I might tinker with overclocking the GPU core a bit and see how much that speeds the folding up. However, I will not be overclocking the GPU memory since the manufacturer decided to leave some of the ram chips bare with no heatsink .

Is the SMP client the best way to go for quad-core CPUs? How does it compare to running 4 instances of the regular client?

Any config tips or parameters that I should add to improve performance?

EDIT: Also, is there some explanation for why "Active clients (within 7 days)" is 1 in my statistics despite me running the GPU client and the CPU client?

-Thanks

[Hufigyed]

Are you using the -bigadv flag?
What is the project number of the WU you are currently folding?
Which project it is makes a pretty big difference in how long it takes.
Folding normal SMPs on my quadcore Athlon 2.9Ghz a frame (1%) takes anywhere from 10-40 minutes depending on the project. My i7-860 @ 3.5Ghz will take anywhere from 5-20 minutes per frame depending on the project.
Folding bigadvs on my i7-860 takes 40-50 min depending on the project.

1 hour per frame looks to be too high for normal SMPs. Do you recall entering -bigadv anywhere during setup?

SMP is better for quads because it will get you more points because you get bonus points based on how fast you finish. You get base points plus bonus.

[andper10]

From the configuration page that I used: ( Folding@home - WinSMPGuide )

4.Configure and use the big setting for "Acceptable size of work assignment" at all times. This 'BigWU' setting is hard coded in the client, so the configuration setting must match. ... Acceptable size of work assignment and work result packets (bigger units may have large memory demands) -- 'small' is <5MB, 'normal' is <10MB, and 'big' is >10MB (small/normal/big) [normal]? This option states a preference for the size of work units downloaded and uploaded to the project servers. Note that all SMP WUs are large in size, so big is the required setting. Bigger units will also have bigger memory requirements. If you run on a slower broadband or dialup internet connection, small is the recommended setting to ease your bandwidth usage and you may want to switch to the CPU client instead.

I chose "big" for the above config option. I'm not sure if that is the same as using the -bigadv flag. I don't see the -bigadv flag in the command prompt window. The current project number my CPU is working on is "Project: 6057 "

(Also, the GPU settings are configured to "allow receipt of work assgiments and return of results greater than 10MB in size.)

[0bit]

From what I have read in another forum, the shader is the only thing that matter for folding. You can leave the memory as is and it won't affect the PPD. Also the ATI/AMD video cards are at a disadvantage compared to the NVidia cards. The GPU clients works better with CUDA(Nvidia) than Open GL(Radeon).

More info

SMP is the way to go if you have a quadcore. I'm not sure what's wrong with your SMP setup. I tried folding on my C2D a while ago and it took about 48 hours to finish. My i7 gets 1% between 13mins and 30mins depending on what else I got going on and my Phenom x4 970 is about the same.

Also don't try the -bigadv unless you have an i7 OCed to around 3.8ghz or a Phenom x6 around the same ghz. You won't make the prefered deadline and won't get the bonus.

The option you chose isn't the -bigadv flag.

[DSCHIEF]

6057 is a standard SMP WU using the A3 core;
On a Q9550 , it should finish in about 12 hrs.
When you ran the first time with the -configonly flag
what % of processor time did you allocate? { I all ways give it 100%}
also give it 100% of detected memory.

The same holds true for setting up the GPU clent, give it 100% of resources. But select Low priority for GPU and select Idle
priority for the SMP client.

Try a copy & paste of the first 20 lines of the log file & paste it here
so we can take a look

Also, to get the bonus points, you'll need to get a Passkey from the stanford site,

http://folding.stanford.edu/English/FAQ-passkey
then re-run with the -config key to add it to your profile.
Bonus points will kick in after you've completed 10 A2 or A3 wu's

[andper10]

I just re-ran the config and it appears that I had set everything up correctly. However, I was running it with the -advmethods flag per a recommendation I read in a forum when I was trying to get it to work originally, so I will try it without that and see if it decreases the processing time (I guess I'll have to wait until I finish this WU to check).

I already had allocated 100% for CPU & Memory for the CPU & GPU clients. I double checked the priorities and they are correct. I got a passkey already. The shortcut I am using has the following flags: -smp -advmethods -forceasm -verbosity 9 (and I just removed -advmethods).

A few other questions:
What is the "prefered deadline"? Is it the same as the "Due time:" listed in unitinfo.txt?
How do you calculate PPD and what exactly does it stand for?

Some lines from the log: (I put (...) in place of information that seemed sensitive)

Note: Please read the license agreement (fah6.exe -license). Further use of this software requires that you have read and accepted this agreement. 4 cores detected 'mpiexec' is not recognized as an internal or external command, operable program or batch file. --- Opening Log file [November 19 00:09:46 UTC] # Windows SMP Console Edition ################################################# ############################################################################### Folding@Home Client Version 6.30 Folding@home - Main ############################################################################### ############################################################################### Launch directory: C:\(...) Executable: C:\(...) Arguments: -smp -forceasm -verbosity 9 -smp [00:09:46] - Ask before connecting: No [00:09:46] - User name: andper10 (Team 13761) [00:09:46] - User ID: (...) [00:09:46] - Machine ID: 1 ***(GPU Machine ID is 2 FYI)*** [00:09:46] [00:09:46] Loaded queue successfully. [00:09:46] [00:09:46] - Autosending finished units... [November 19 00:09:46 UTC] [00:09:46] + Processing work unit [00:09:46] Trying to send all finished work units [00:09:46] Core required: FahCore_a3.exe [00:09:46] + No unsent completed units remaining. [00:09:46] Core found. [00:09:46] - Autosend completed [00:09:46] Working on queue slot 02 [November 19 00:09:46 UTC] [00:09:46] + Working ... [00:09:46] - Calling '.\FahCore_a3.exe -dir work/ -nice 19 -suffix 02 -np 4 -che ckpoint 15 -forceasm -verbose -lifeline 9976 -version 630' [00:09:46] [00:09:46] *------------------------------* [00:09:46] Folding@Home Gromacs SMP Core [00:09:46] Version 2.22 (Mar 12, 2010) [00:09:46] [00:09:46] Preparing to commence simulation [00:09:46] - Ensuring status. Please wait. [00:09:56] - Assembly optimizations manually forced on. [00:09:56] - Not checking prior termination. [00:09:56] - Expanded 1764909 -> 2251569 (decompressed 127.5 percent) [00:09:56] Called DecompressByteArray: compressed_data_size=1764909 data_size=22 51569, decompressed_data_size=2251569 diff=0 [00:09:56] - Digital signature verified [00:09:56] [00:09:56] Project: 6057 (Run 0, Clone 154, Gen 149) [00:09:56] [00:09:56] Assembly optimizations on if available. [00:09:56] Entering M.D. [00:10:02] Using Gromacs checkpoints [00:10:04] Resuming from checkpoint

Thanks for all the help!

[Hufigyed]

I always run it with the -advmethods flag and it never takes that long, I don't think thats the problem. Yeah 60 min per frame with 6057 WU on that cpu is slow.

PPD is "points per day" as you may have seen, you get points based on how much work you do for the f@h project and people compete with each other trying to get more points. stats It is also deent a way of verifying that everything is working right by seeing how many PPD other people are getting with the same or similar hardware.
You can calculate your PPD here. Folding@home SMP2 Bonus Point Calculator

What is the -forceasm flag? I don't recall seeing that before.

Originally Posted by 0bit F Also don't try the -bigadv unless you have an i7 OCed to around 3.8ghz or a Phenom x6 around the same ghz. You won't make the prefered deadline and won't get the bonus.

Actually RAM makes a bigger difference here. A capable CPU is required of course. But I ran bigadvs on my i7 860 at stock speed(2.8Ghz) and made the deadline. Right now I have it at 3.5Ghz and easily make the deadlines.
I have 8GB of 1600Mhz CAS 7 ram.

[andper10]

The -forceasm flag was supposed to force assembly optimizations in case the program tried to mistakenly shut them off. I removed it though. Now I just have "-verbosity 9 -smp" added to the shortcut.

How exactly do I find someone with a Q9550 CPU to compare my results with using the provided stats link?

I am now trying project 6701 . It is averaging about 15 minutes & 18 seconds per frame @ stock (2.83 Ghz) with no GPU client running. According to the bonus point calculator, that yields about 4170.94 PPD. Does that sound more in line with what a Q9550 should be getting? I will continue to tinker with it as I have spare time.

[Hufigyed]

I usually search Google for my cpu and the project number and usually there is some forum where someone has mentioned time. I find quite a bit of stuff at Folding Forum • Index page .

Just a quick seartch finds this "TPF on 6701 on a Q9550 @ 3.8 GHz running Win7: 9:58" on this page.

So that's quite a bit faster than yours. Are you also using this machine for other cpu intensive stuff?
What I would probably do is start a thread at foldingforum.org and post your TPFs on different WUs as you do them and they should be able to tell you if it's normal for that CPU or not.

I think you should be able to do more than 4000 PPD with that CPU but I have no actual experience with that CPU. Also some WUs will work out to a lower PPD than others.

[andper10]

I am still running it stock at the moment. I will put the overclock back and see if it is in line with 9:58 TPF.

Thanks

[Hufigyed]

Oops my bad. Stock is 2.8Ghz and the post I found had the CPU at 3.8Ghz. I just didn't notice that the numbers were different. So never mind my above post, your CPU is probably performing as it should.

The only thing with OCing for this type of project is to make sure that you get no errors. No errors is more important that finishing faster. So make sure you can run fully stable in Prime95 with 0 errors for at least 24hours but if you'll be folding 24/7 like I am, you may want to test for an even longer continuous time.

[flanzig1]

If your running an SMP client and a GPU client, the GPU client will take one of the cores leaving 3 cores for the SMP.
I have an Q9600 running SMP and it is doing about 20minutes a frame for core 6701.
My other quad is running a GPU client and 3 regular clients on the remaining cores.

[andper10]

I got the overclock back up to 3.7Ghz and it was stable in Prime 95 for 30 minutes without going over the max temp threshold. I will run Prime95 for longer after this WU sends itself off. I am getting about 4:40 - 4:50 TPF for project 6061 without the GPU client running, which should yield 8780.19 to 9254.73 PPD. I looks like everything is working correctly now.

Is there a points calculator for the GPU version so I can see if it will be more beneficial to run it along with the CPU version vs. just running the CPU version?

P.S. I am probably upgrading to an i7 soon so more points for the team!

Thanks for all the help everyone!